My findings are striking: for every eight days (including weekends) since school cancellations began, a county tends to have one less death per 100,000 people. For every nine days a ban on gatherings over 500 people has been in place, there’s one less death per 100,000 people. These policies work. But the correlation flips for bans on gatherings of fifty people or for stay-at-home orders. For every two weeks a stay-at-home order is in place, the death rate rises by one person per 100,000. For bans of gatherings of fifty people, it’s every eleven days.
Because I controlled for how long it’s been since the first county death, this is probably not being driven by harder-hit places choosing to adopt tougher policies. But just to check, I also added a control for the deaths per 100,000 people in each county on March 31. This does reduce the size and significance of effects for each policy, but not the relationship between them. Nor are my findings changed by excluding the New York City metro area, or adding or removing a handful of other variables related to climate, industry or occupation mix, rurality, etc. All the underlying data can be downloaded publicly.
The only US-based academic study empirically linking lockdowns to lower deaths is a recent economics paper identifying California’s lockdown as the reason for its lower death rate. The problem with this paper is that the authors find that the lockdown began to reduce California’s deaths just five days after being implemented. The effect is too early to derive from the supposed cause.